[atomic-primops] Bugs and build problems on 7.10 RC1 about haskell-lockfree HOT 10 CLOSED

Thanks, I'm taking a look. By the way, what version of cabal-install was this?

from haskell-lockfree.

Thanks for looking into it. I just upgraded cabal-install:

jberryman /tmp/atomic-primops-0.6.1 » cabal --version
cabal-install version 1.20.0.6
using version 1.20.0.3 of the Cabal library 

Aside from the base constraint, this seems like it is some sort of weird cabal issue or maybe something wonky with my install, but I don't know.

from haskell-lockfree.

(Oh, and a bit off topic, but I've been trying to figure out if there have been any interesting developments vis a vis atomic-primops in 7.10? Making them inline primops or something?)

from haskell-lockfree.

With cabal install head (1.22.0.0), and 7.10 RC1 on linux, the atomic-primops library installs and passes tests for me on Linux, e.g. when I run with:

cabal-1.22.0.0 install -w ~/opt/ghc-7.10-rc1/bin/ghc --run-tests ./ ./testing/

(I'm waiting for 7.10 to build on my laptop to test on Mac as well. But I went ahead and pushed a patch relaxing the base constraint in any case.)

However, I did see 2/107 test failures on this linux machine that need to be investigated.

casmutarray1: [Failed]
Unboxedbasic1_incrCounter: [Failed]
incrCounter returns the NEW value
expected: 10
 but got: 0    

I can't produce that error with GHC 7.8.3, so it seems to actually be a regression in GHC 7.10. @tibbe, do you have any idea why your new (more efficient) versions of the primops might be failing these tests? (Or might be different in some way that exposes a failing in the test harness!)

from haskell-lockfree.

Huh, in some combinations of messing with sandboxes I can get this error:

Linking dist/dist-sandbox-c87b52cc/build/test-atomic-primops/test-atomic-primops ...
dist/dist-sandbox-c87b52cc/build/test-atomic-primops/test-atomic-primops-tmp/Test.dyn_o: In function `rlYU_info':
(.text+0x4dc): undefined reference to `atomizuIpTkfAN9tvQCCr8X2lp3Yl_DataziAtomicsziInternal_zdfShowTicket1_closure'

But I haven't reproduced the same error as you got with cabal-1.20.0.6. Instead, I'm currently having problems like this:

cabal-1.20.0.6
cabal-1.20.0.6 install ./ ./testing/ --run-tests -w ~/opt/ghc-7.10-rc1/bin/ghc
ghc: ghc no longer supports single-file style package databases (dist/dist-sandbox-c87b52cc/package.conf.inplace) use 'ghc-pkg init'

from haskell-lockfree.

Small update. If I build without a sandbox, I get to the same point with cabal-1.20.0.6 as I did with the HEAD version. That is, it installs fine and 2/107 tests fail.

In any case, I've gone ahead and uploaded the relaxed bounds version to Hackage, because well, at least most of the library works ;-). But we need to hurry to fix that regression!

from haskell-lockfree.

I factored out the failure of Unboxedbasic1_incrCounter into issue #41.

I still need to look at the casmutarray1 failure.

from haskell-lockfree.

Thanks for looking into this so quickly. FWIW the main issue I was having was unrelated to atomic-primops (I describe what I had to do to get a working 7.10 here). I'll close this since you've released the updated base constraint version, and you have a separate ticket for the issue you uncovered.

from haskell-lockfree.

Sorry, changed my mind; I'll let you close this when you're ready :)

from haskell-lockfree.

Sure, either way. Currently, I see this ticket as for the cabal incompatibility you cited. So I changed the title to that effect. Currently I have not successfully reproduced that using cabal-install 1.20.0.6 / Cabal 1.20.0.3. Can you get that same error reproducably?

I also added #42 for the other failed test and a "ghc-7.10" milestone for all issues related to GHC 7.10 support.

from haskell-lockfree.